home page of Janmejaya Panda, systems programmer.
my twitter my music my github email
LLVM/Clang contributor since 2025. I work on clang semantic analysis hardening, llvm-libc functions, and RISC-V codegen research. I like working on things where a single misplaced bit can ruin someone's entire Tuesday.
source code [on github]:
Aion: LLVM optimization debugger that answers the only question compiler engineers actually care about: "why didn't it optimize?" decodes pass remarks, multi-context IR analysis, covers vectorization, PGO, devirtualization -- basically the stuff LLVM doesn't tell you unless you ask very nicely (C++ / LLVM / YAML)
Lain-OS (codenamed Abyss): experimental kernel I'm building because I got annoyed at how operating systems manage resources. 10-layer architecture, hierarchical NeuroVFS, a temporal fracture scheduler which is exactly as chaotic as it sounds, and a predictive shell that uses markov chains to guess what you'll type next (it's disturbingly accurate) (C / RISC-V ASM)
aetherfm: GTK+ file manager that doesn't suck. fast, minimal, actually respects UNIX philosophy unlike certain other file managers I could name (C / GTK / GLib)
Aether Allocator: single-header C allocator. zero overhead. eliminates leaks, double-free, dangling pointers. the kind of thing you write after you've spent one too many nights debugging someone else's malloc crimes (C)
inter-cpp: bytecode interpreter with a custom opcode dispatch mechanism I'm unreasonably proud of (C++17)
P.A: process analyzer. finds and harvests zombie and orphan processes via syscall instrumentation. yes, "harvests" is the right word (C++)
rune: pure Haskell classification library. human-readable rules, no neural network black-box nonsense. if you can't explain why a decision was made, you don't have a classifier, you have a religion (Haskell)
krrs: what happens when you try to write a kernel in Rust. turns out the borrow Checker has opinions about interrupt handlers (Rust)
bootloader-asm: x86 bootloader in pure assembly. real-mode to protected-mode. the 47 most important bytes of any computer's life (x86 ASM)
VoidAllocator: memory allocator tuned for bulk integer allocations. turns out when you stop pretending everything is a generic object, things get fast (C++)
bgkernel: a real working kernel model. bootloading, interrupts, the whole stack. yes it actually boots (C++ / ASM)
QuantumTeleportMAUI: quantum teleportation simulator. I wanted to understand quantum computing so I built a teleporter. no particles were harmed (Q# / C#)
Aion is the primary active project. the core loop is deceptively simple but the implications are deep:
// the question every compiler engineer loses sleep over
auto remarks = parser.getRemarks();
for (auto& R : remarks) {
if (R.isMissed()) {
analyzeInhibition(R);
}
}
I write: C, C++, Rust, Haskell, Go, Java, Python, x86/RISC-V Assembly
I break: Linux kernels, memory allocators, compiler pipelines, bytecode VMs
I use: LLVM, CMake, Docker, GDB, Git, PyTorch, TensorFlow
I don't just use computers; I dissect them. from the Linux Kernel to the LLVM IR, driven by a deep obsession with how software talks to hardware. currently architecting Lain-OS (experimental kernel), Aion (LLVM debugger), and Aether (safe memory models). considers sleep(8) optional.
I'm obsessed with movies, especially how they look. I spend a lot of time studying directors like Kubrick, Kurosawa, Tarantino, and Coppola. Every shot they take feels like it was handled with so much care.
There are a few shots that just stay in my head. Like the way Kurosawa handles movement in Seven Samurai, or that entire movie Barry Lyndon which was shot entirely in candlelight with vintage lenses. Even just the tension in a Tarantino opening—it’s just perfect. I try to bring that same level of care to my code.
I listen to a lot of shoegaze, rock, and dreampop. I've always loved bands that can create a massive wall of sound where everything just blurs together. Most of my spotify is basically just a collection of reverb and echoing vocals.
I tend to like games that are actually hard and force you to really pay attention. Mostly FromSoftware stuff. I think that "git gud" mindset is actually pretty useful for building kernels—you just keep trying until it finally clicks. If it's too easy, I usually get bored pretty fast.